token limit on fireworks deepseek v3 editor

[gminorcoles]

Issue

Hi,
I am having trouble with token limits in /architect mode using fireworks ai deepseek v3 as the editor. this happens quite often using r1 as the main model and for the first time it happened with sonnet as the main model. I have model settings borrowed from another user who says they use fireworks successfully. the token max being zero makes me think its fireworks issue? but they are not helpful.

$ 0.0800 26,661 tokens total

Aider v0.72.3
Main model: claude-3-5-sonnet-20241022 with diff edit format, infinite output
Weak model: fireworks_ai/accounts/fireworks/models/deepseek-v3
Git repo: .git with 280 files
Repo-map: using 4096 tokens, auto refresh

Model fireworks_ai/accounts/fireworks/models/deepseek-v3 has hit a token limit!
Token counts below are approximate.

Input tokens: ~23,784 of 0 -- possibly exhausted context window!
Output tokens: ~1,951 of 0 -- possibly exceeded output limit!
Total tokens: ~25,735 of 0 -- possibly exhausted context window!

To reduce output tokens:

• Ask for smaller changes in each request.
• Break your code into smaller source files.
• Use a stronger model that can return diffs.

To reduce input tokens:

• Use /tokens to see token usage.
• Use /drop to remove unneeded files from the chat session.
• Use /clear to clear the chat history.
• Break your code into smaller source files.

model settings
[
{
"name": "fireworks_ai/accounts/fireworks/models/deepseek-r1",
"edit_format": "diff",
"weak_model_name": "fireworks_ai/accounts/fireworks/models/deepseek-v3",
"use_repo_map": true,
"send_undo_reply": false,
"lazy": false,
"reminder": "user",
"examples_as_sys_msg": true,
"cache_control": false,
"caches_by_default": true,
"use_system_prompt": true,
"use_temperature": false,
"streaming": true,
"editor_model_name": "fireworks_ai/accounts/fireworks/models/deepseek-v3",
"editor_edit_format": "editor-diff",
"remove_reasoning": "think",
"extra_params": {
"max_tokens": 160000
}
},
{
"name": "fireworks_ai/accounts/fireworks/models/deepseek-v3",
"edit_format": "diff",
"use_repo_map": true,
"reminder": "sys",
"examples_as_sys_msg": true,
"extra_params": {
"max_tokens": 128000
}
}
]

model metadata
"fireworks_ai/accounts/fireworks/models/deepseek-r1": {
"max_tokens": 160000,
"max_input_tokens": 120000,
"max_output_tokens": 20480,
"input_cost_per_token": 0.000008,
"output_cost_per_token": 0.000008,
"litellm_provider": "fireworks_ai",
"mode": "chat"
},
"fireworks_ai/accounts/fireworks/models/deepseek-v3": {
"max_tokens": 128000,
"max_input_tokens": 100000,
"max_output_tokens": 20480,
"input_cost_per_token": 0.000001,
"output_cost_per_token": 0.000001,
"litellm_provider": "together_ai",
"mode": "chat"
},

Version and model info

No response

[FelixLisczyk]

I've also been experiencing unusual token limits with DeepSeek via Fireworks.ai since this morning:

Model fireworks_ai/accounts/fireworks/models/deepseek-v3 has hit a token limit!
Token counts below are approximate.

Input tokens: ~7,791 of 128,000
Output tokens: ~1,886 of 8,192
Total tokens: ~9,677 of 128,000

Here is my Aider version and model config:

Aider v0.72.3
Model: fireworks_ai/accounts/fireworks/models/deepseek-v3 with whole edit format
Git repo: ../.git with 90 files
Repo-map: using 4096 tokens, files refresh

[lpmwfx]

I also experience the infinite mode does not work. Aider v0.73.0

Model: deepseek/deepseek-chat with diff edit format, prompt cache, infinite output

The warning:
Your estimated chat context of 150,916 tokens exceeds the 128,000 token limit for
deepseek/deepseek-chat!

I expected Aider to cut off exceeding chat log for models that support infinite.

[akaihola]

I'm using

Aider v0.72.4.dev42+g99e3c1ce
Model: fireworks_ai/accounts/fireworks/models/deepseek-r1 with architect edit format
Editor model: openrouter/anthropic/claude-3.5-sonnet:beta with diff edit format

and I`m getting

Model fireworks_ai/accounts/fireworks/models/deepseek-r1 has hit a token limit!
Token counts below are approximate.

Input tokens: ~9,441 of 0 -- possibly exhausted context window!
Output tokens: ~1,908 of 0 -- possibly exceeded output limit!
Total tokens: ~11,349 of 0 -- possibly exhausted context window!

To reduce output tokens:
- Ask for smaller changes in each request.
- Break your code into smaller source files.
- Use a stronger model that can return diffs.

To reduce input tokens:
- Use /tokens to see token usage.
- Use /drop to remove unneeded files from the chat session.
- Use /clear to clear the chat history.
- Break your code into smaller source files.

[paul-gauthier]

@akaihola what version of aider? You may want to aider --upgrade

@lpmwfx there's nothing aider can do if you hit your provider's context window limit. Infinite output helps with the output limit, but the context window size is enforced by your provider.

[github-actions]

I'm labeling this issue as stale because it has been open for 2 weeks with no activity. If there are no additional comments, I will close it in 7 days.

Note: A bot script made these updates to the issue.

[github-actions]

I'm closing this issue because it has been stalled for 3 weeks with no activity. Feel free to add a comment here and we can re-open it. Or feel free to file a new issue at any time.

Note: A bot script made these updates to the issue.